Slashdot Mirror

Examples of human-competitive results using genetic programming (i.e. the algorithm we refer to as Evolution):
http://www.genetic-programming...

Read this and this

Chris Anderson
has foreseen the most profound change since the age of reason. Man has
reached the point where his "understanding" can impede evolution. It is
time to concede that some processes may be beyond our
comprehension.

Research in the area of Artificial General Intelligence provides a
crystal clear demonstration of the problem. A half century of research
has led to "intelligent" data mining and voice response systems and
very little else.

However, Koza, Fogel
and many others have observed evolutionary computation machines
creating solutions to real world problems. In some cases these
are patentable solutions beyond previous human achievement, and some of
them defy understanding.

Unless you have unlimited funding and lots of time, it's not necessary
to understand why every complex solution works. It may not even be
possible.

A million MRI's of functioning brains are not likely to result in any
Lisp program for AGI, so the search for AGI seems to be coming full
circle back to the "baby bootstrap". Even Ben Goertzel
is looking to virtual babies to mine the clouds.

Like others who have managed to see beyond the horizon, Anderson will
be widely misunderstood. He is not rejecting scientific method, he is
simply showing us its limitations.

Following years? How about now?

http://www.genetic-programming.com/humancompetitiv e.html

http://www.human-competitive.org/

(2006 results aren't posted yet...)

I was at the GECCO06 conference (Genetic and Evolutionary Computation COnference) when the Human Competitive awards were handed out. The first place winner went to a guy whu evolved an oscillator that used HALF as many capacitors and resistors than the industry standard one. The second place winner evolved input parameters to Schrodinger's equations that enabled him to model a certain chemical reaction 10 and in some cases 1000 times faster than the earlier best result in the literature. And the rest of the entries were fascinating as well.

http://www.genetic-programming.org/hc2006/entrants _table.html

The 1st and 2nd place entries were #5 and #6 respectively.

I don't understand why some American christians feel so threatened by the evolution theory. If I can use evolution, including truly random mutations and all, to make something, then surely so can God? Evolution is not magic, it's a mechanism. A tool, if you will, and it can be used to make something that does exactly what you want it to, but is to complex to design by hand.

Take a look at Genetic Programming to see what I'm talking about.

The most important argument of the Intelligent Design crowd seems to be that life is too complex to be the result of a random process, but these kind of random processes produce amazingly (even ridiculously) complex solutions all the time. They just don't really understand what they're talking about. Evolution may not be experimentally provable in biology, but it has already been proven in Artificial Intelligence.

And it doesn't even conflict with a belief in God. What more do you want?

Read up a bit more on http://www.genetic-programming.org/Genetic Programming. It's more powerful and flexible than standard genetic algorithms, but also a lot more complex and unpredictable. Possibilities are endless, and they've only begun to discover just how many kinds of problems you can solve with evolution this way.

Also an interesting idea is to use genetic algorithms to set up the parameters for a neural network. Tuning NN isn't easy, so why not let evolution do it for you?

mcv.

You were probably talking about biological evolution, and the silly idea that it usurps the role of a creator.

"There are now 36 instances where genetic programming has automatically produced a result that is competitive with human performance, including 15 instances where genetic programming has created an entity that either infringes or duplicates the functionality of a previously patented 20th-century invention, 6 instances where genetic programming has done the same with respect to a 21st-centry invention, and 2 instances where genetic programming has created a patentable new invention.".

Now the computational power of these computers is faily meager. I think the largest cluster applied so far has been 1000 pentium 350's. The "computational" power of a population of species is massive. If quantum computers can be developed, and genetic programming algorithms can be written in such a way that takes advantadge of the properties of quantum machines, we *really* will be entering a new era in humanity (however there is no indication this is possible or not possible, I am just speculating)

If they can evoluate, why not try to show them how to find the best solution on a given computer program?

It's already been done. Genetic algorithms is the evolution of a particular solution to achieve the best results, such as finding the best solution to the travelling salesman problem. Travelling to 100 cities in the best possible order to minimize the distance travelled, this would take 100! (that's factorial, not just an exclamation) calculations to search through the entire solution space. Using a genetic algorithm, you can evolve solutions -- in effect, borrowing snippets of cities from one solution and breeding them with another solution and seeing if it's more fit than the others.

Genetic programming is the application of genetic algorithms to actual computer code to produce an optimal program.

Very nice example. This is why I think genetic/evolutionary algorithms are so interesting and will definately be used a lot more in the future.

On a similar note, there's genetic programming, where the building blocks are pieces of syntactically correct code("if", "then", "for"), and your fitness function is how fast the resultant program can solve a particular problem.

Its interesting to see that given a problem like sorting, genetic programming will come up with the same solutions that we humans have already thought of. So that must mean there are no algorithms that are more efficient than the ones we have already thought of.

I wonder if maybe one day this will solve the P=NP problem. :-)

Cheers,

Costyn.

*** Mild spoilers, if you haven't read at least the first two books in the series ***

Rama talks about biots, which is short for biological robots. It's a combination of organic matter with mechanical/electronic parts. The crew captures one and takes it apart to find a combination of biological based batteries (think: electric eel), electronic parts (for recording, visualization, etc.), and other parts. The idea is furthered in the subsequent book, Rama II where the Octospiders manage to engineer biological creatures to do work for them. For example, a dragonfly-type machine can be built to do a videorecording of a scene. When it's done recording, it uploads its data to a central location and, in exchange, is rewarded for doing the assignment correctly by receiving energy or food (or karma points, whatever).

There will be an eventual convergence of biology and computers. And I'm not talking about simulated biology via genetic algorithms/programming Preliminary progress is being made, but I await the day I can plug in a 1 terabyte hard drive into my brain!

[knee-jerk] Burn Em! Bury em alive! [/knee-jerk]

One interesting thing to note about patents is that they can be used as a yardstick to measure human intelligence in such a way that it can be compared literally against machine intelligence. This is, in fact, one of Alan Turings' proposed measurements of artificial intelligence. In fact, machines have already designed systems that meet or exceed the capabilities of human patents. From the linked page:

Fourteen of the results are competitive with human-produced results. Ten infringe on previously issued patents or duplicate the functionality of previous patents in novel and creative ways.

Just thought I'd bring a slightly positive note to what is already becoming Caldera Flamefest 2003

A simple proof of evolution is to look at genetic programming (for example here, here and here).
Just look at the classic example of ants collection food. It is beautifully described in John R. Koza's great books (1, 2 and 3) on the subject.
Just imagine adding a fermone layer to freeciv and let the random search for a superior player begin.

You wouldn't get results in this millenium if you did that. First of you want to make sure that you only generate valid code, and from what I've heard bytecode is a bit of a bitch to do by hand so generating java files which you compile is probably a better idea. (Or just use parameters which you tweak, that is the usual way of doing genetic programming.)

You need to feed the generator with a lot of domain info too, otherwise it will just produce garbage. (Even if it is compilable garbage.)

If you are interested in this stuff check genetic-programming.org.

I know that some guy used genetic programming to generate bots for Quake1. They were really stupid though. (He had a sort of blog where he told what had happened lately.) He got bots which would kill you right away if you circle strafed to the left. If you circle strafed to the right they would be completely lost. It's fun but not very usable.

You are right about the GA, but this is closer to Genetic Programming than GA. GA evolves the 'answer', but GP evolves the 'solution' to the answer.

There's a difference, GA is much easier to program than GP and is usually much faster. Example of a good candidate problem for to use GA to solve would be the travelling salesman problem, while GP would find a method to solve a problem.

While fundamentally pretty similar, GP is slightly more complicated. You have to deal with issues such as program over growth during evolutions, which doesn't happen in GA.

http://www.genetic-programming.org/ is a good source to learn more about GP.

I took a class on GA and GP while in college, very interesting stuff. :)

Err, just for clarification, not _all_ genetic algorithms need to have sex (aka exchange of genetic material). In reality what is being described is Evolutionary Programming or sometimes as genetic programming.

Out of interest, I wrote a simple program which creates a random file. [...] It turns out on my systems, and with code ( please don't mock my newbie code :), improve it if you will ), never ever produced any file that could executed properly, let alone do something useful. [...] I just wanted to try and put it in perspective, how unlikely it is that life was created randomly.

If you want to redo your experiment using reasonable methods, check out the field known as "genetic programming".

Natural selection isn't the only process acting, either; there is also self-organization. For a very fascinating theoretical argument (backed up by computer simulations) as to why life might not only be not unlikely, but inevitable, check out Stuart Kauffmann's At Home in the Universe .

So it is STUPID to think that life, which is far more complex than a computer program, happened only by chance.

Check out genetic programming. Automatic programming via genetic algorithms.

The Father of GP (John Koza) may disagree with me - he runs genetic-programming.org and more or less invented the field. He's also known for his vigorous defences of GP: anybody know of real applications?

A somewhat more complete description of GP can be found at Genetic-programming.com.

How can something designed, programmed, and worked on hard by humans become better than the capacity of the human(s)' mind/intelligence that designed it?

There are quite a few examples of endeavours in which the human mind designed things that outsmarted it. Although it is controversial to do it, you simply can't say that Deep Blue does not play chess better than any human that designed it.

But the example I always like to give when such discussions are held is that of genetic programming. Genetic programming is an area of evolutionary computation that tries to achieve automatic programming. It basically uses GA techniques to evolve programs. There are reported cases of results in which the program outsmarted human beings quite nicely. One great book in the subject is Evolutionary Design by Computers, a collection of texts and papers in the subject, edited by Peter Bentley.

All in all, most AI criticisms seem to degenerate in anthropocentric pseudo-arguments. Another good book to be read is Dreyfus' What computers (still) can't do. Dreyfus gives good reasons for why AI may be far from the present, but does so without (for the most part, at least) resorting to the argument that "I'm human and want to be the only smart being here". It is interesting that AI criticism may be the last island of anthropocentrism. First, the Sun does not go around the earth, but otherwise. Then, me and that disgusting worm are made of the same genetic stuff. Now, a bunch of transistors beats me at chess and wants to think? Then again, this is just me.

The links are here for the paranoid:
http://www.genetic-programming.org
http://www1.fatbrain.com/asp/bookinfo/bookinfo.asp ?theisbn=155860605X&vm=
http://www1.fatbrain.com/asp/bookinfo/bookinfo.asp ?theisbn=0262540673&vm=

Carlos
Semper ubi sub ubi

Actually I heard that there are 10^12 neurons in the human brain that are able to perform 10^3 pulses per second (leaving us with 10^15 tics/sec or a petaflop). This is what John Koza (Mr. Genetic Programming) calls a Brain Second (or 1 BS).

Interestingly he has found in his research that there seems to be a problem convergence in his evolving of circuits where the processor time needed to evolve a circuit takes roughly 1 BS computation.

He's worked his way up to a 1000 node Beowulf cluster for his work over the years and has seen this 1 BS ratio maintained (i.e. the faster his computative guns, the lower the total evolution time for an optimal result).

He charted the growth of his testbed and found that in something like 2004 a 1 BS per second should be achievable in industry and privately available by 2010.

All in all very interesting stuff. Actually it should all be lain out in his upcoming book (appropriately titled Genetic Programming IV) and should make for some killer reading...

Genetic programming has been going on quite a while. I don't understand why people get so excited when an old idea gets implemented in hardware. Doing something in hardware instead of software is just an optimization. It runs faster, but it still does pretty much the same thing.

It is very possible that all programming in ten to twenty years with be done by artificial intelligence engines utilizing genetic algorthms.

You may have to retrain in some new field if you do not make it rich with your new internet startup

There have already been some radical advances in this are that are downright scary. The programs look like nothing any human would right, but they work.

there are plenty of links discussing this here at genetic-programming.org

One big plus to this is that this has real potential to make Microsoft truely obsolete, when everyone can grow their own software.

It is very possible that all programming in ten to twenty years with be done by artificial intelligence engines utilizing genetic algorthms.

You may have to retrain in some new field if you do not make it rich with your new internet startup

There have already been some radical advances in this are that are downright scary. The programs look like nothing any human would right, but they work.

there are plenty of links discussing this here at genetic-programming.org

One big plus to this is that this has real potential to make Microsoft truely obsolete, when everyone can grow their own software.

If you are interested in genetic programming take a look here for more info.

--

Last I heard, John Koza (at Stanford) was in the process of building a 1,000-Pentium Beowulf-Style Cluster Computer for Genetic Programming.

Koza has been doing very cool things with GP for a long time, ever since he was at Michigan working with John Holland.